Mary Kay Trims Servers, Taps Windows Mainframe

Kregg Jodie knows exactly how well commodity hardware handles added demands.

As CIO of Mary Kay Inc., the $1.3 billion direct seller of cosmetics and beauty products, he built the company’s entire electronic commerce effort on servers made by Compaq and Dell and software from Microsoft. Hardware that automatically balanced demands made it possible for Jodie’s team to grow the company’s online operations incrementally and take advantage of the relatively low cost of Windows administration and development—up to a point.

Jodie hit that point this year. Throwing more servers and related storage and networking hardware at increasing requests for information and more plentiful transactions just didn’t cut it any more.

The added demands were swamping performance—on peak days, as many as 30,000 orders would come in from the Internet. With each order consisting of as many as 50 separate individual transactions, with their own associated credit card numbers and inventory look-up, that amounted to about a million transactions—”mostly coming at the end of the day,” says Jodie. At that load level, he says, there was no way to eliminate the “database bottleneck” by just adding more commodity hardware.

So Mary Kay ditched servers and moved its electronic business onto a Unisys mainframe. With that move, Mary Kay signaled that it believes, like Corporate Express, the Atlanta-based utility Southern Company, and corporate travel services provider Galileo International, that as sites grow, centralized computing makes more sense on the Web.

In Mary Kay’s case, the move is major but not as revolutionary as it might have been—the mainframe runs Windows. The ES7000 is the first server hardware designed specifically for Microsoft’s Windows 2000 Datacenter Server operating system, applying mainframe switching technology to Intel processors.

“This Unisys hardware is really interesting,” says Galen Schreck, enterprise infrastructure analyst at the Forrester Research firm. “The real question is where it fits in, versus racks full of smaller servers.”

Schreck says that while big multiprocessor systems are useful for consolidating applications built to take advantage of multiple processors—like databases and transactional systems—there are many parts of data center automation that can’t. “A lot of data centers rely heavily on scripts and batch operations that aren’t even multithreaded,” he says, meaning they can’t break up their processes into multiple tasks that can be run at the same time. For these applications, there’s no performance boost to be had in a multiprocessing system.